Re-assessing the filtering rationale

I want to check the rationale and details of our previous filtering philosophy as Jun had the feeling that we removed too many cells. We also removed too many genes, e.g. highly tissue specific genes such as myl7 are not part of the final data set any longer, which is something that we need to go back on.

sce <- readRDS(paste0(data_dir, "sce_ZF-allSamples-Integrated_2022-01-12_noCounts.rds"))
rownames(sce) <- scater::uniquifyFeatureNames(rowData(sce)$ID, rowData(sce)$Symbol)
sce$old_labels_coarse <- ifelse(grepl("Macrophage", sce$label), "M0",
    ifelse(grepl("T-cells", sce$label), "Tcell",
    ifelse(grepl("Fibroblast", sce$label),"Fibroblast",
        ifelse(grepl("Cardiomyocyte", sce$label), "CM", 
            ifelse(grepl("Endocardium", sce$label),"Endocard.",
                ifelse(grepl("Epicardium", sce$label), "Epicard.",
                    ifelse(grepl("Ery", sce$label), "Ery",
                    ifelse(grepl("eutroph", sce$label), "Neutrophils",
                        ifelse(grepl("delta", sce$label), "delta",
                        ifelse(grepl("EC_", sce$label), "EC",
                            ifelse(grepl("ndothelial", sce$label), "EC", sce$label)))))))))))
sce$old_labels_coarse <- gsub("-cells", "cell", sce$old_labels_coarse)
sce$old_labels_coarse <- gsub("cells$", "", sce$old_labels_coarse)

p1 <-scABC2::plot_reducedDim_from_sce(sce,
    which_reddim = "UMAP", color_by = "hbba1_ENSDARG00000097238",
    exprs_values = "logcounts",
    size_by = .5, alpha = .3, set_colors = FALSE,
    remove_rug=TRUE,
    add_cell_info = c("Tissue","post.surgery")) +
    facet_grid(.~Tissue) + scale_color_viridis_c(direction=-1) +theme(legend.position = "bottom") + ggtitle("Ery marker")

p2 <-dittoSeq::dittoBoxPlot(sce,
    var = "hbba1_ENSDARG00000097238", 
    group.by = "cluster_k200", split.by = "Tissue",
    boxplot.lineweight = 0.1, jitter.size = .2, jitter.width=0.5)

p1 + p2

• Clusters 7, 9, 13 show high Ery-markers.

plot_reducedDim_from_sce(sce[, as.character(sce$cluster_k200) %in% c("7","9","13")],
    which_reddim = "UMAP", 
    color_by = "cluster_k200",
    add_cell_info = "Tissue") +
    facet_wrap(~Tissue)+ ggtitle("Clusters 7, 9, 13 (high Hbba1)")

• cluster 13 is most likely the true ery cluster because it’s shared across all four tissues

plot_reducedDim_from_sce(sce, which_reddim = "UMAP", 
    color_by = list(title="Ery label?", 
        result = ifelse(sce$old_labels_coarse == "Ery", TRUE, FALSE)))

plot_reducedDim_from_sce(sce, which_reddim = "UMAP", 
    color_by = list(
        title="Ery label?", 
        result = ifelse(sce$old_labels_coarse == "Ery", TRUE, FALSE)),
    add_cell_info = c("Tissue")) + facet_wrap(~Tissue)

• most erys that were identified by reference label transfer were found in the liver
• cluster 7 in the heart probably primarily represents contaminated CMs

–> check (a) CM markers, (b) liver markers [hepatocytes, cholangiocytes]

Ery vs CM

goidt <- scABC2::make_long_dt(sce, exprs_values = "logcounts",
    features = c("ttn.2","tnni1b", # CM
        "col1a1a", #epicardial
        "hbba1_ENSDARG00000097238", # ery
        "cela1.6", #exocrine pancreas
        "tfa","fabp10a", #hepatocytes,
        "lgals2b", #cholangiocytes
        "colec11", # HepSC
        "itga2b", # megakaryo
        "slc4a1a" # ery
        ),
    include_metaData = c("cluster_k200", "old_labels_coarse","Tissue","Sample","post.surgery","detected","subsets_mito_percent"))

goidt <- dcast(goidt, cell + Tissue + post.surgery + cluster_k200 + Sample + subsets_mito_percent + detected + old_labels_coarse ~ feature_name,  value.var = "logcounts")
setnames(goidt, "hbba1_ENSDARG00000097238", "hbba1")

ggplot(goidt, aes(x = ttn.2, y = hbba1, color = cluster_k200)) + 
    geom_point(size = 1.5, alpha = .2, shape = 1) +
    facet_grid(Tissue~post.surgery) +
    scale_color_manual(values = ABCutilities:::fx.get_palette_ABC("paired_pal")) +
    guides(color = guide_legend(override.aes = list(size = 3, alpha =1, shape = 16) ) )

plot_reducedDim_from_sce(sce[, as.character(sce$cluster_k200) %in% c("7","9","13")],
    which_reddim = "UMAP", 
    color_by = "ttn.2", exprs_values = "logcounts",
    add_cell_info = "Tissue", set_color = FALSE) +
    facet_wrap(~Tissue)+ ggtitle("CM marker", subtitle="Clusters 7, 9, 13 (high Hbba1)") + scale_color_viridis_c(direction = -1, option = "magma")

Ery vs. Liver cells

ggplot(goidt, aes(x = colec11, y = hbba1, color = cluster_k200)) + 
    geom_point(size = 1.5, alpha = .2, shape = 1) +
    facet_grid(Tissue~post.surgery) +
    scale_color_manual(values = ABCutilities:::fx.get_palette_ABC("paired_pal")) +
    guides(color = guide_legend(override.aes = list(size = 3, alpha =1, shape = 16) ) ) +
    ggtitle("Cholangiocyte marker")

• cholangiocytes don’t seem to have a lot of hbba1

ggplot(goidt, aes(x = fabp10a, y = hbba1, color = cluster_k200)) + 
    geom_point(size = 1.5, alpha = .2, shape = 1) +
    facet_grid(Tissue~post.surgery) +
    scale_color_manual(values = ABCutilities:::fx.get_palette_ABC("paired_pal")) +
    guides(color = guide_legend(override.aes = list(size = 3, alpha =1, shape = 16) ) ) +
    ggtitle("Hepatocyte marker")

• fabp10a also often found in pancreas cells (cluster 9; labeled as “acinar cells”)
  • according to this paper, fabp10a should be very specific to the liver
• there are plenty of cells with high fabp10a in the liver sample that don’t show very high hbba1
  • makes me want to flag cluster 7 liver cells

dittoSeq::dittoBoxPlot(sce[,sce$Tissue=="Liver"], var = "detected",
    group.by = "Sample", split.by = "cluster_k200", 
    boxplot.lineweight = .1, jitter.size = .5, main = "Liver samples only")

plot_reducedDim_from_sce(sce[, sce$Tissue == "Liver"],
    which_reddim = "UMAP", 
    color_by = "subsets_mito_percent", exprs_values = "logcounts",
    add_cell_info = c("cluster_k200","Tissue"), set_color = FALSE) +
    facet_wrap(~Tissue)+ ggtitle("Liver cells only") + 
    scale_color_viridis_c() + facet_wrap(~cluster_k200) +
    theme(legend.position = "bottom")

There seem to be two mito-populations in cluster 7.

ggplot(goidt[Tissue == "Liver"],
    aes(x = fabp10a, y = hbba1, color = subsets_mito_percent)) + 
    geom_point(size = 1.5, alpha = .2, shape = 1) +
    facet_grid(post.surgery~cluster_k200) +
    scale_color_viridis_c()+
    guides(color = guide_legend(override.aes = list(size = 3, alpha =1, shape = 16) ) ) +
    ggtitle("Hepatocytes only") + theme(legend.position = "bottom")

ggplot(goidt[Tissue == "Liver"],
    aes(x = fabp10a, color = hbba1, y = detected)) + 
    geom_point(size = 1.5, alpha = .2, shape = 1) +
    facet_grid(post.surgery~cluster_k200) +
    scale_color_viridis_c(direction=-1, option = "magma")+
    guides(color = guide_legend(override.aes = list(size = 3, alpha =1, shape = 16) ) ) +
    ggtitle("Hepatocytes only") + theme(legend.position = "bottom")

• cluster 7 liver cells can probably be removed

Erys vs acinar/exocrine markers

ggplot(goidt, aes(x = cela1.6, y = hbba1, color = cluster_k200)) + 
    geom_point(size = 1.5, alpha = .2, shape = 1) +
    facet_grid(Tissue~post.surgery) +
    scale_color_manual(values = ABCutilities:::fx.get_palette_ABC("paired_pal")) +
    guides(color = guide_legend(override.aes = list(size = 3, alpha =1, shape = 16) ) ) +
    ggtitle("Acinar cell marker")

• cluster 9 cells indeed seem to also express highest levels of exocrine cell marker cela1.6 –> let’s keep it

Summary

• keep cluster 13 = erythrocytes
• re-label cluster 7 & heart cells = cardiomyocytes
• remve cluster 7 from all other samples (predominantly liver with low numbers of genes detected)